Abstract: In facial expression recognition, the feature extraction methods such as Gabor combined with Local Binary Pattern （LBP） and histograms for dimension reduction are more localization methods, but the LBP is of poor robustness and low precision for recognition, and the computational complexity and the dimension of characteristics are still high for histograms. Compared with LBP, Center-Symmetric Local Binary Pattern （CS-LBP） is more robust, but its description for facial expression texture is insufficiently detailed. This paper proposes a new feature extraction method based on Gabor combined with improved CS-LBP. The improved CS-LBP is addition of two-valued center-symmetric local binary pattern. The method uses Gabor to extract feature, and the final recognition feature is the superposition of two different features calculated by two methods. The dimensions of the final recognition feature are effectively reduced by Discrete Cosine Transform （DCT）. The proposed method is tested by using JAFFE facial expression database. The result shows that the method can effectively improve the recognition accuracy.

Key words: facial expression recognition, Gabor wavelet, addition of two-valued center-symmetric local binary pattern, discrete cosine transform, feature dimension reduction

摘要： 在表情识别中Gabor结合局部二值模式（LBP）的特征提取方法以及直方图统计降维虽然是较为局部化的方法，但LBP鲁棒性较差，识别精度不高，而且使用直方图统计来区分表情，其计算复杂度和特征维数依旧较高。中心对称局部二值模式（CS-LBP）与LBP相比具有较好的鲁棒性，但其对表情纹理细节的描述仍不够详细。因此提出基于Gabor结合改进的CS-LBP即二值叠加中心对称局部二值模式（二值叠加CS-LBP）的特征提取方法。用Gabor提取特征，同时用两种计算方式提取两个特征值并叠加，作为最终识别的特征；并通过离散余弦变换（DCT）降维，有效降低表情的特征维数。在JAFFE表情库中实验验证了该方法能有效提高识别精度。

关键词: 表情识别, Gabor小波, 二值叠加中心对称局部二值模式, 离散余弦变换, 降维